374. Requirements for ML Systems
Requirements The requirements for an ML system vary from use case to use case. However, most systems should have these 4 characteristics. Here is a brief overview. Reference Designing Machine Learning Systems
Kyosuke
373. Knowledge Distillation
The Challenges While training large models helps improve state-of-the-art performance, deploying such cumbersome models fail to meet performance benchmarks at the time of inference on real-world test data. Knowledge distillation helps overcome these challenges by “distilling” the knowledge in a…
372. Time Required For Model Production
Return on investment in ML depends a lot on the maturity stage of adoption. Here is an overview of how long it takes for a company to bring a model to production by Algorithmia. Reference: Designing Machine Learning Systems
371. Framing ML Problems
Identifying the Task How you frame an ML problem can significantly affect the difficulty of solving it. It’s important to appropriately identify the task type of your ML problem to avoid such situations. Let’s say we want to predict what…
370. ML System Development Process
Process Developing ML systems are iterative and in most cases a never-ending process. Here is a simplified representation of what that iterative process looks like.
369. Batch Prediction VS Online Prediction
Difference The terms “online prediction” and “batch prediction” can be confusing. Both can make predictions for multiple samples or one sample at a time and can make predictions asynchronously(when online prediction leverages a real-time transport to send prediction requests to…
368. ML Systems
Components When people say “Machine Learning”, they usually think of “Machine Learning Algorithms”, but actually, algorithms are only a small part of the entire system. Here is an overview of what is necessary to compose a Machine Learning System. Reference…
367. Transfer Learning VS Fine-Tuning
Difference Transfer learning and Fine-tuning both replace the final layer to make predictions for a different task. However, while transfer learning fixes all the layer previous to the final output layer, fine-tuning initializes the weights with the pre-trained weights but…
366. Categorical Label Mask For Semantic Segmentation
Preparing Your Dataset For Training When training a multi-class segmentation model, loss functions such as the CrossEntropyLoss() in Pytorch require the mask to be categorically labeled. As in the image above, the RGB mask (middle) is properly segmented but all…
365. Applying Data Augmentation Using Albumentation
When applying data augmentation for computer vision models, it is important to apply the same augmentation in the same order for both the input and target image. Here is one way to implement using Albumentation and Pytorch. Implementation Today I’ll…